Files, cleanup and troubleshooting
Keep each processing lineage together
New ERPERF chains keep source recordings and timing files in the study and
write derived products below erperf/<recording>/. A named branch writes to
its branch subfolder. Reused legacy root-level chains are not moved silently.
Important companion files include sensor configurations, processing settings, ICA metadata/decisions, epoch condition files, artefact decisions and trial selection manifests. Copying just the largest data file is not a portable processing checkpoint.
Bundle existing ERPERF artifacts… on Flow control deliberately migrates legacy subject products. Preview the sources/destinations. Raw recordings, timing and study-wide files are protected; collisions prevent migration. Downstream subject-owned products can also be included. Keep a backup.
Prune intermediate filters
Prune intermediate filter files offers eligible middle recordings of the configured chain, preserving the source, broadband ICA correction, final recording and metadata. Check the preview and preserved prerequisites.
An incomplete final recording prevents pruning. Branch filter intermediates may be retained conservatively when recovery from an inherited source cannot be established. This is preferable to guessing which files are expendable.
Delete outcomes deliberately
Delete outcomes from selected step onward… is a destructive operation. Choose the batch rows, branch and starting stage, then inspect the exact preview.
The grouping scope applies when deleting only final averaging products. Starting earlier invalidates every grouping that depends on those prerequisites. The optional folder checkbox removes branch/subject bundle folders after they become empty; it is not permission to recursively delete arbitrary data.
Earlier checkpoints, protected sources and unrecognized files are not a license to assume the whole folder can be removed. Deleting a parent result can also make a child branch’s inherited checkpoint stale.
Deletion requires confirmation and is permanent, not moved to Trash. Do not run cleanup while processing is active. Restore deleted products from backup or recalculate them from valid prerequisites.
Diagnose a stopped run
| Message or symptom | What to check |
|---|---|
| Missing timing/tag | Run MARKER or restore the timing file; select an existing tag |
| Filter/sample-rate mismatch | Match configured filters to the actual rate; resample explicitly if needed |
| Incompatible ICA training checkpoint | Restore the original settings or include preparation/decomposition in the rerun |
| Changed parent checkpoint | Recalculate the branch from its fork |
| Unreadable sensor configuration | Restore the matching channel order/geometry, not an arbitrary cap file |
| Missing filter provenance | Read the compatibility warning; rebuild if exact provenance is required |
| Zero accepted trials or failed balancing | Inspect threshold decisions, timing constraints and the selected profile |
| Missing/incomplete outputs | Inspect the file-specific write error, free space and permissions |
| Too many matches | Narrow the search to one branch, processing suffix and product type |
Some missing/empty derived configurations can be recovered from a validated geometry-preserving prerequisite. A [RECOVERED] log entry describes that operation. Recovery is not performed by replacing a nonempty configuration with an arbitrary one, and cannot infer geometry after a spatial transform.
Check the file-specific error before rerunning. A batch configured to continue after errors may finish while individual recordings remain incomplete. Use the outcome reports and keep normal study backups.